Akash Gupta Is Teaching Warehouses to Think in Motion

Akash Gupta once landed in the United States with three suitcases. One held his clothes. The other two were packed with electronic components, servo motors, circuit boards, soldering irons, bolts, nuts and enough tools to suggest that the luggage itself was an engineering project. Thomas Chance, whose marine-technology company took in Gupta and his BITS Pilani teammate Samay Kohli around 2009, remembered the inventory years later. Gupta spent nights moving between code, electronics, motors and gears. The striking detail was not the clutter. Chance said the things Gupta built worked.

The suitcase ratio is a useful map of Gupta's mind: make room for the machinery first. Long before he became co-founder and CEO of GreyOrange, he was a schoolboy in Uttar Pradesh who started coding in class six with GW-BASIC. The next year, he borrowed his sister's copy of Let Us C. Then came 3D animation in 3ds Max and Maya, plus a parallel life as his school's 100-meter champion. By the time he entered BITS Pilani in 2008 to study mechanical engineering, he had spent years making things move on a screen. He wanted the physical world to catch up.

A presentation by Kohli's robotics team gave that frustration a body. The team was building AcYut, one of India's early indigenous humanoid robots. Gupta saw an object that demanded everything at once: mechanics, electronics, embedded systems, servers, sensors and software. He joined. On some days, the group worked for 18 hours. They designed machines from scratch, traveled to competitions across countries and learned the merciless feedback loop of physical engineering. A program can be elegant and still send a robot in the wrong direction.

A company hiding inside a workshop

Gupta and Kohli did not begin with a polished warehouse thesis. Their early business grew out of teaching robotics workshops at colleges, supported by ₹5 lakh from savings earned during internships. The workshops taught them scale, but also showed them what kind of company they did not want to run. Curriculum management and motivating instructors were not their strengths. Building was.

After several white-label projects, they wrote rules for choosing an industry. The problem had to be global. The solution had to be meaningfully different. It also had to combine software and hardware, ideally with simple, elegant machinery and complex software behind it. They considered robots for oil and gas maintenance, machines for railway inspection, and supply-chain automation. Warehousing survived the filter.

Warehouses were a good place for an engineer who distrusted paper understanding. They are diagrams made stubbornly real. A shelf has weight. An aisle has width. A worker has a route. An order has a clock. One delayed tote can alter the value of every decision around it. The founders' Butler system brought together mobile robots, storage units, pick-put stations and software carrying the business logic of inventory. It was not one product so much as four products required to behave like one.

The early company had to manufacture its own credibility along with its machines. India offered few experienced engineers who could cross hardware, embedded systems and software, and a young startup could not simply recruit a veteran bench. GreyOrange hired people with strong analytical ability and trained them for the work. Gupta valued humility alongside raw problem-solving skill because failure was built into the learning curve. Prototyping presented another obstacle. Suitable custom shops were scarce, so the team created machining and electronics facilities of its own. This made growth more expensive, but it shortened the distance between an idea, a broken part and the next attempt. The approach echoed the AcYut lab: keep the disciplines close enough that feedback travels quickly. It also helped answer a commercial question facing an Indian hardware company selling abroad. Customers did not buy the founders' college story. They bought equipment that had to perform against established international suppliers, shift after shift.

From the robot to the result

For most of GreyOrange's first decade, Gupta was the technical founder in the partnership. “I was always the person who went super deep into technology,” he said of the division of labor. He served as chief technology officer and later took on product leadership. Kohli, the original CEO, concentrated on company building and the broader product direction. Their college roles had scaled into executive ones.

But the center of gravity gradually moved. GreyOrange had launched proprietary robotics hardware in 2012. Over time, the larger opportunity became coordination: software capable of assigning work to different, job-specific robots, including machines made by other vendors, using near-real-time information. GreyMatter, the company's fulfillment orchestration platform, gave that idea a name. The warehouse was no longer a stage built around a single machine. It was a changing system of orders, inventory, workers and varied robotic agents.

Gupta became CEO in April 2023, with Kohli moving to the board. The transition formalized a broader remit that had already been forming. Gupta would lead strategy while retaining an unusual intimacy with research, product and the floor. Later that year, GreyOrange closed $135 million in Series D growth financing. In his account of 2024, the company deployed its gStore retail platform across more than 1,900 stores in 38 countries while extending warehouse automation into additional industries.

The retail piece brought its own education. Gupta has written that before building further in retail technology, he worked full time for three months at an H&M store in Beverly Hills. He knew supply-chain systems; the store was new ground. What stayed with him was the gap between putting products on shelves and shaping the experience after a customer walks in. For a founder steeped in abstractions, the response was characteristically physical: take the job, stand on the floor, watch the system misbehave.

AI that has to touch the floor

Gupta now describes the next chapter as physical AI. The phrase can invite science-fiction theater, but his version begins with less glamorous questions. How should hundreds of robots find collision-free routes? Where should inventory live? Which worker or machine should take the next task? How should a carton be packed? Each is a hard optimization problem. They are also connected, changing one another while the building keeps operating.

The practical objective is not mathematical perfection. It is a continuous search for better answers using heuristics, optimization, machine learning, human expertise and feedback from the real environment. This also changes what counts as success. A warehouse can report 100 percent robot uptime while failing to ship what customers ordered on time. The machine metric is locally impressive and operationally beside the point. Gupta's preferred scorecard starts with fulfillment outcomes.

That line captures the distance between language intelligence and embodied operations. Warehouses and stores generate sensor data, location changes, congestion, substitutions and human improvisation. Gupta has discussed building world models for these spaces: give a model the current state and possible actions, then ask what the environment will look like afterward. Reverse the question and the model can suggest actions that move the system toward a target state. He calls the work early, somewhere between research and product, which is refreshingly precise for a field crowded with premature certainty.

His story has a clean arc only in retrospect. Coding led to animation. Mechanical engineering led to a humanoid. A humanoid led to workshops. Workshops led to warehouse robots. Robots led to orchestration software. Warehouses led back out to stores. The connecting thread is not a single technology. It is a habit of following an operational problem until another discipline becomes necessary.

That habit also explains the writable walls and desks reported in Gupta's old Gurugram office, crowded with formulas, diagrams and lists. The surface did not matter as much as the impulse: make the system visible, then work on it. Somewhere between the ceiling fan he could draw but not identify and a warehouse that can revise thousands of decisions in real time, Gupta learned to close the distance between knowing and making. The next machine he is building is not a single robot. It is the judgment of the whole floor.